Hi,
Glad that helped!
-
Having categorical predictors doesn’t change the choice of sampler (PyMC3 chooses the best one for you in most cases anyway), but you have at least two ways of specifying the priors for them: either with indicator variables or with index variables. The latter is usually easier to specify priors and is also more scalable to cases with many categories as well as hierarchical models. You’ll find examples of index variables for categorical predictors in chapter 11 of Rethinking 2 – here is the port to PyMC3.
-
You can definitely do PP plots with ArviZ. The best is that you sift through the API functions.
Hope that helps 